Original Link: https://www.anandtech.com/show/405
AMD Athlon Buyer's Guide - Part 2: Overclocking
by Anand Lal Shimpi on November 9, 1999 8:48 PM EST- Posted in
- Guides
Be sure to read Part 1 of our Athlon Buyer's Guide
When we first took a look at a final production version of AMD's Athlon CPU on a fully functional Slot-A motherboard, one of our initial reactions was, "Hey, you can't overclock this thing." A quick call to AMD revealed that they had taken a different approach to controlling the clock multiplier and voltage settings of the Athlon CPU than the market had been used to. Instead of allowing for the two options to be controlled and adjusted on the motherboard as they have been in the past, AMD placed the only method for controlling the clock multiplier and core voltage on the Athlon processor card itself.
This manipulation is done by a physical modification of a combination of 16 resistor slots on the processor card. A number of Taiwanese websites published the first instructions on the actual manipulation of a few of these resistors to increase the clock multiplier of the processor, but Dr. Thomas Pabst of Tom's Hardware Guide was actually the first to produce a complete guide on all of the available settings, including the settings for changing the core voltage. The guide is very well written and contains all of the background information necessary to know how one can physically overclock the Athlon, so take a look at it if you are truly interested.
Since the release of that article, many individuals that are a little more than skilled with their soldering irons have taken it upon themselves to make a few bucks off of the modification. The fact of the matter is that when AMD released the Athlon back in August, they could already hit 750MHz in their own labs, so it makes perfect sense to expect that the Athlon is at least a decent overclocker. At the same time, the higher clock speed Athlons are out of the price range of many, while the Athlon 500 weighs in at a very competitive and reasonable price. What do these two events have in common? Take an Athlon 500, overclock it to 650MHz, and you now have a 650MHz Athlon for around $250.
Unfortunately, the modification process is very dangerous and should be reserved for those with quite a bit of experience with delicate soldering techniques. This is where the idea of capitalism comes in: at an added premium, a number of companies are offering pre-overclocked Athlon CPUs that operate at the higher speeds but cost lower than the higher clocked part would go for. Is there anything wrong with this? Not at all, as long as the vendor discloses that these CPUs are in fact overclocked and there are risks involved.
Some individuals have even taken it to the next level and have actually outfitted their modified Athlon CPUs with a set of dip switches that can control the clock multiplier and core voltage so that the control is placed back in the hands of the user. We were lucky enough to obtain an evaluation sample from one of these individuals, Mark Sorensen, the owner and operator of Trinity Micro. Trinity Micro is currently offering a modified Athlon 500 for $310, as well as the opportunity to have your own CPU modified for $125.00 + shipping and handling. While the pricing may seem quite steep, we had tremendous success with our evaluation Athlon 500, as you're about to see. Note that this potential does not apply to Trinity Micro's Athlon units alone, as they are all using the same chips, just different methods of modifying them in order to achieve the common goal: taming and overclocking the world's most robust x86 CPU.
The yields on the Athlon have been incredible; they are high enough that some users are reporting that their lower clock speed Athlon CPUs feature L2 cache chips that would be normally found on the higher speed CPUs. One such case was that of our Athlon 500 evaluation unit from Trinity Micro which featured 3.6ns L2 cache, normally reserved for 600MHz Athlon CPUs.
With the L2 cache capable of hitting the 300MHz required at a 600MHz clock speed, it came to no surprise that our Athlon 500 evaluation unit made it up to 600MHz without even the slightest hiccup. By increasing the core voltage to 1.75v we were able to get the Athlon to POST at 800MHz. Keep in mind that this is from a 500MHz CPU and we disabled the L2 cache in order to test the core itself.
Unfortunately, the problem with this method is that disabling the L2 cache significantly decreases performance in most applications and the raw clock speed increase then only serves as a benefit if you're running a 3D game or another similar application that is not L2 cache dependent. Luckily, there is a way to adjust the L2 cache divider by a method similar to the one used to adjust the clock multipliers and it allows you to change the L2 cache divider from the default /2 (1/2 * clock speed) to /3 or even /1 for a 1:1 L2 cache to core clock speed ratio.
While the latter option isn't really feasible since we would need at least 2ns L2 cache in order to even hit the 500MHz requirement of an Athlon 500 with a 1/1 L2 cache divider, the 1/3 multiplier makes quite a bit of sense. If your L2 cache is the limiting factor with the overclocking potential of your CPU (this can be determined by turning off the L2 cache, and seeing how far your CPU will overclock, then turning it back on and seeing how far it will go then) it may be worth it to settle for the slower 1/3 speed L2 cache but at the higher core clock speed.
So how do you control the L2 cache divider on the Athlon? We already know all of the other settings for clock multiplier and voltage from Tom's Article and below are the settings for the various L2 cache frequency dividers:
Using R103, R104, R106 and R107 located on the back of the Athlon card, the settings are as follows:
L2 Cache Ratio |
R103 | R104 | R106 | R107 |
1/1 | OFF | ON | OFF | ON |
1/2 | OFF | ON | ON | OFF |
1/3 | ON | OFF | OFF | ON |
1/5 (Off) | ON | OFF | ON | OFF |
What are the tradeoffs of the lower L2 cache clock speed versus the higher core clock speed? That's what we're here to find out, but first let's take a look at the options offered to us by Trinity Micro's modified Athlon.
The Test
Windows 98 SE Test System |
|
Hardware |
|
CPU(s) |
Trinity Micro AMD Athlon 500 - Modified |
Motherboard(s) | ASUS K7M 1.04 |
Memory | 128MB PC133 Corsair SDRAM |
Hard Drive | IBM Deskstar 22GXP 22GB Ultra ATA 66 HDD |
CDROM | Phillips 48X |
Disk Controller | VIA On-board Ultra ATA 66 Controller |
Video Card | NVIDIA GeForce 256 SDR Reference Board |
Ethernet | Linksys LNE100TX 100Mbit PCI Ethernet Adapter |
Software |
|
Operating System |
Windows 98 SE |
Video Drivers | NVIDIA Reference Detonator 3.53 drivers |
Benchmarking Applications |
|
Business |
BAPCo SYSmark 98 |
Gaming | idSoftware Quake 3 Test 1.08 (OpenGL) |
Starting with Overall System Performance, measured by BAPCo's SYSMark 98, the precedence seems to be given to the setups with the higher overall clock speed. The Athlon overclocked to 750MHz with a 250MHz L2 cache (1/3 * 750 = 250MHz) outperforms the Athlon 700 with its 350MHz L2 cache and the Athlon 700 with 233MHz L2 cache outperforms the Athlon 650 with 325MHz L2 cache.
The performance of overclocked Athlons is a very positive sign for owners of slower CPUs whose L2 cache cannot cope with speeds much greater than 300MHz. Imagine purchasing an Athlon 500 that could overclock to 700MHz easily, with the only necessary tweak being that you reduce the L2 cache speed to 233MHz from your original 250MHz. The performance benefit is definitely worth it. But does this trend continue as we explore the individual applications SYSMark 98 uses to derive its score?
SYSMark 98 |
|
Athlon 500/166 |
223 |
Athlon 500/250 |
232 |
Athlon 550/183 |
241 |
Athlon 550/275 |
249 |
Athlon 600/200 |
258 |
Athlon 600/300 |
264 |
Athlon 650/217 |
274 |
Athlon 650/325 |
280 |
Athlon 700/233 |
284 |
Athlon 700/350 |
292 |
Athlon 750/250 |
304 |
All scores are Time in Seconds, Lower is Better
As we've come to learn and expect, L2 cache speed has a weak effect on image editing applications. There is a difference between the 1/2 and 1/3 L2 cache dividers, but the L2 cache performance improvement is really more dependent on the size of the L2 cache than the speed. For this reason, an Athlon 600 with a 200MHz L2 cache (1/3) outperforms an Athlon 550 with a 275MHz L2 cache (1/2) and so on and so forth.
Photoshop 4.0.1 |
|
Athlon 500/166 |
81.29 |
Athlon 500/250 |
79.62 |
Athlon 550/183 |
74.42 |
Athlon 550/275 |
73.07 |
Athlon 600/200 |
71.44 |
Athlon 600/300 |
69.22 |
Athlon 650/217 |
66.98 |
Athlon 650/325 |
66.93 |
Athlon 700/233 |
63.85 |
Athlon 700/350 |
62.9 |
Athlon 750/250 |
60.32 |
A similar situation is present in the CorelDRAW test; the results mirror those of the Photoshop test from above.
CorelDRAW 8.0 |
|
Athlon 500/166 |
180.02 |
Athlon 500/250 |
175.23 |
Athlon 550/183 |
165.33 |
Athlon 550/275 |
161.86 |
Athlon 600/200 |
153.37 |
Athlon 600/300 |
149.1 |
Athlon 650/217 |
143.9 |
Athlon 650/325 |
139.64 |
Athlon 700/233 |
135.97 |
Athlon 700/350 |
129.45 |
Athlon 750/250 |
128.95 |
All scores are Time in Seconds, Lower is Better
3D Rendering Applications are also more dependent upon L2 cache size rather than speed, so changing the divider to 1/3 from 1/2 doesn't have such an adverse effect. Performance in Bryce 2 drops a couple of percent on the Athlon 700 as the L2 cache speed drops from 350MHz to 233MHz, but even with a 233MHz L2 cache, the Athlon 700 is slightly faster than the Athlon 650 with a 325MHz L2 cache. This is the case in the majority of 3D rendering application tests.
Bryce 2 |
Elastic Reality 3.1 |
Extreme 3D 2 |
|
Athlon 500/166 |
104.96 |
88.26 |
73.52 |
Athlon 500/250 |
100.21 |
84.46 |
72.47 |
Athlon 550/183 |
96.77 |
82.3 |
68.38 |
Athlon 550/275 |
92.49 |
78.99 |
66.41 |
Athlon 600/200 |
89.64 |
75.29 |
63.42 |
Athlon 600/300 |
86.7 |
73.79 |
62.87 |
Athlon 650/217 |
82.96 |
71.86 |
60.53 |
Athlon 650/325 |
79.06 |
68.47 |
58.83 |
Athlon 700/233 |
77.88 |
69.5 |
56.28 |
Athlon 700/350 |
76.46 |
64.17 |
56.91 |
Athlon 750/250 |
73.64 |
63.32 |
53.15 |
All scores are Time in Seconds, Lower is Better
Business/Office Applications are where the faster L2 cache is preferred over the slower L2 cache with the faster CPU clock speed. Most business/office applications are small enough to fit within the L2 cache of the Athlon and thus a faster L2 cache is preferred to the faster CPU clock speed. At the same time, keep in mind that you don't need to have the world's fastest CPU just to run business/office applications. If you're using your computer for more than just business applications, the performance gains in other areas will make up for the loss in business apps.
These last few tests seem to offer a small tradeoff between L2 cache speed and CPU clock speed, so the deciding factor comes down to how much of a sacrifice you're willing to give up in some areas in order to gain in others.
Excel 97 |
PowerPoint 97 |
Word 97 |
|
Athlon 500/166 |
107.63 |
67.46 |
87.93 |
Athlon 500/250 |
99.49 |
63.79 |
83.13 |
Athlon 550/183 |
101.87 |
60.19 |
81.36 |
Athlon 550/275 |
94.89 |
59.85 |
78.32 |
Athlon 600/200 |
96.55 |
57.64 |
76.71 |
Athlon 600/300 |
90.93 |
55.28 |
74.91 |
Athlon 650/217 |
87.1 |
53.11 |
71.39 |
Athlon 650/325 |
84.09 |
52.82 |
69.36 |
Athlon 700/233 |
88.8 |
53.75 |
69.68 |
Athlon 700/350 |
83.54 |
50.62 |
67.93 |
Athlon 750/250 |
79.55 |
48.46 |
65.4 |
All scores are Time in Seconds, Lower is Better
Naturally Speaking 2.02 |
Netscape 4.05 |
OmniPage Pro 8.0 |
Paradox 8.0 |
|
Athlon 500/166 |
126.1 |
85.62 |
93.17 |
62.42 |
Athlon 500/250 |
122.07 |
81.13 |
91.01 |
60.77 |
Athlon 550/183 |
120.95 |
80.64 |
84.41 |
58.34 |
Athlon 550/275 |
115.41 |
77.37 |
83.58 |
56.64 |
Athlon 600/200 |
113.17 |
75.21 |
78.42 |
54.48 |
Athlon 600/300 |
111.57 |
74.18 |
78.97 |
52.31 |
Athlon 650/217 |
108.38 |
72.8 |
74.07 |
52.89 |
Athlon 650/325 |
107.88 |
71.19 |
72.64 |
50.82 |
Athlon 700/233 |
104.35 |
69.75 |
69.13 |
49.96 |
Athlon 700/350 |
103.22 |
67.21 |
68.59 |
47.97 |
Athlon 750/250 |
100.3 |
66.72 |
65.31 |
47.24 |
Premiere 4.2 |
XingMPEG Encoder 2.1 |
|
Athlon 500/166 |
45.68 |
86.38 |
Athlon 500/250 |
43.46 |
84.5 |
Athlon 550/183 |
41.39 |
81.66 |
Athlon 550/275 |
40.73 |
79.03 |
Athlon 600/200 |
38.59 |
74.8 |
Athlon 600/300 |
37.6 |
75.13 |
Athlon 650/217 |
35.31 |
72.24 |
Athlon 650/325 |
34.46 |
72.69 |
Athlon 700/233 |
33.36 |
71.84 |
Athlon 700/350 |
33.53 |
72.95 |
Athlon 750/250 |
31.04 |
67.13 |
Gaming performance is another area where the higher clock speed means more than a faster L2 cache. Our Quake 3 tests can be treated much like the 3D rendering tests from before; they are dependent primarily upon raw FPU performance, which scales tightly with CPU clock speed rather than L2 cache frequency.
As the resolution goes up, the performance gap decreases, but the trend remains the same -- clock speed over L2 cache speed.
At 1024 x 768 we seem to hit the ceiling with the GeForce, which eliminates most of the performance differences among the CPUs.
Quake 3 Arena Test demo1.dm3 |
|||
- |
640 x 480 |
800 x 600 |
1024 x 768 |
Athlon 500/166 |
99.8 |
97 |
77.4 |
Athlon 500/250 |
104.7 |
100 |
77.8 |
Athlon 550/183 |
104.8 |
100.6 |
77.8 |
Athlon 550/275 |
110.6 |
104.2 |
78.1 |
Athlon 600/200 |
111.1 |
105.5 |
78 |
Athlon 600/300 |
116 |
107 |
78.3 |
Athlon 650/217 |
116.7 |
108 |
78.3 |
Athlon 650/325 |
119 |
109.9 |
78.4 |
Athlon 700/233 |
121.2 |
109.9 |
78.4 |
Athlon 700/350 |
123.3 |
111.3 |
78.4 |
Athlon 750/250 |
125.4 |
111.5 |
78.4 |
Summary
The overclocking options offered by the Athlon are tremendous, and they help the performance user tweak every last bit of power out of their system. Simply increasing the clock multiplier is the most effective way to obtain a performance increase, but when limits such as the L2 cache come into play then, at times, this isn't possible. The 1/3 L2 cache frequency divider is a viable option as well, since the performance drop is negligible. This option is also an indication that the future Athlon CPUs with a 2/5 L2 cache divider won't sacrifice too great of a performance loss since the 2/5 divider will still place the L2 cache at a higher frequency than the 1/3 divider does. A full speed L2 cache would still be preferred, but that's some time away, at least for now.
As for possible Athlon overclocking options, Trinity Micro's solution is a fairly decent one. The only real problem we found with it was that it prevents the use of certain coolers, such as the MC1000, which we reviewed a while back. If anything would make the perfect addition to a modified Athlon, it would be a peltier cooler capable of driving temperatures well below zero.
It's only a matter of time before do-it-yourself kits emerge for the Athlon, either by offering tips on soldering or by taking advantage of the Athlon's unique feature connector at the top of the PCB. It's about time that AMD leaked the data sheets on that connector.
Part 3 of this guide will focus on finding the best video card for your new Athlon system, so keep your eyes peeled for that addition to this series.